基于SSA-XGBoost的岩性密度测井曲线环境校正方法OA

Affected by the combined extreme engineering conditions including intense compressive in-situ stress and nitrogen underbalanced drilling,deep clastic formations are prone to borehole collapse,which severely distorts litho-density logging curves.Conventional chart plate methods and multiple regression approaches fail to reconstruct the true formation logging responses.To address this issue,this paper proposes a method for environmental correction of litho-density log curves based on the sparrow search algorithm(SSA)for hyperparameter optimization of extreme gradient boosting(XGBoost).The core workflow adopts SSA to conduct global adaptive optimization on critical XGBoost hyperparameters(i.e.,the number of decision trees,learning rate and maximum tree depth),and constructs an SSA-XGBoost petrophysical nonlinear mapping model.Multilayer perceptron(MLP)neural network,random forest(RF)and unoptimized raw XGBoost are selected as benchmark models.Systematic comparative analysis is carried out from three dimensions:convergence characteristics of error loss curves,quantitative accuracy metrics on validation sets,and comprehensive field application performance in actual wells,which verifies that the SSA-XGBoost model delivers the optimal correction performance.Quantitative research results indicate that:①The litho-density of the Jurassic Ahe formation in the northern Kuqa tectonic belt presents a well-coupled controlling relationship with natural gamma ray,acoustic transit time and deep resistivity.By introducing petrophysical constraints into the feature space of machine learning,the organic integration of formation physical mechanisms and data-driven modeling is realized.②A coefficient of determination of 0.88 is obtained for the SSA-XGBoost model on the test dataset,alongside a root mean square error of 0.027 and a mean absolute error of 0.019.Relative to the baseline XGBoost model without SSA optimization,the prediction precision is elevated by 6.02%.Field verification on well Dx1 demonstrates that the litho-density curves corrected by SSA-XGBoost achieve the highest consistency with measured logging data from non-hole-enlarged intervals,with mean relative error as low as 0.47%.③Global adaptive optimization of key hyperparameters(tree count,learning rate,maximum tree depth)via SSA effectively eliminates the defect that manual hyperparameter tuning of standard XGBoost easily falls into local optima,and significantly enhances the model's noise immunity and generalization capacity.It is concluded that the environmental correction method for litho-density log curves proposed in this paper provides an efficient and robust solution for the environmental correction of litho-density logs in deep clastic reservoirs,and offers reliable technical support for petrophysical research and fluid property identification.